How do I see what the browser is doing?

You can use the 'screenshot' command to see the current state, or run the CLI with the '--headed' flag to open a visible browser window for debugging.

Does it support parallel browsing?

Yes, it supports independent browser sessions using the --session flag, allowing you to automate multiple websites simultaneously without state interference.

Can I use it to scrape data behind a login?

Absolutely. You can script the login sequence using 'fill' and 'click' commands, then use 'snapshot' or 'get' commands to extract data from protected pages.

Is this different from the Playwright MCP?

Yes. While Playwright MCP uses direct tool integrations, agent-browser uses a Bash-based CLI interface, which is often faster for one-off automation tasks and simpler to debug via terminal output.

How does ref-based selection work?

The skill generates accessibility snapshots that assign temporary IDs (like @e1) to interactive elements, allowing Claude to click or fill them precisely without needing complex CSS selectors or XPaths.

Agent Browser Automation

Name: Agent Browser Automation
Author: udecode

byudecode

•

389

•

浏览器自动化

Automates headless browser interactions and web data extraction using a ref-based CLI optimized for AI agents.

Agent Browser Automation provides a specialized CLI interface for Claude to interact with web pages using Vercel's agent-browser tool. Unlike traditional automation frameworks, it utilizes a unique ref-based selection system (@e1, @e2) derived from accessibility snapshots, making it significantly more accurate for LLMs to identify and interact with DOM elements. It supports comprehensive navigation, complex form filling, high-resolution screenshot capture, and structured data scraping across parallel sessions, serving as a robust alternative to Playwright for rapid, bash-driven web workflows.

主要功能

01High-fidelity web scraping with support for text and JSON output

02389 GitHub stars

03Full browser control including navigation, clicks, and form filling

04Capture viewport or full-page screenshots and generate PDF exports

05Ref-based element selection using interactive accessibility snapshots

06Parallel session management for simultaneous multi-site automation

使用场景

01Visual UI audits and screenshot generation for automated testing

02Automated form submission and multi-step login sequence execution

03Dynamic web scraping for competitive data collection or analysis

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add udecode/better-convex agent-browser

For use in Claude.ai and ChatGPT

主要功能

01High-fidelity web scraping with support for text and JSON output

02389 GitHub stars

03Full browser control including navigation, clicks, and form filling

04Capture viewport or full-page screenshots and generate PDF exports

05Ref-based element selection using interactive accessibility snapshots

06Parallel session management for simultaneous multi-site automation

使用场景

01Visual UI audits and screenshot generation for automated testing

02Automated form submission and multi-step login sequence execution

03Dynamic web scraping for competitive data collection or analysis

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add udecode/better-convex agent-browser

For use in Claude.ai and ChatGPT